![Arxiv:1802.07403V3 [Math.AG] 15 Jun 2020 Udeo Rank of Bundle and Theorem Theorem Have: We Then Theorem Iesosadabtaycaatrsi L,Tm .] Ntecs O Case the in 0.1]](http://data.docslib.org/img/01d7b6a5a216d84aee9b0ca809118274-1.webp)
arXiv:1802.07403v3 [math.AG] 15 Jun 2020 udeo rank of bundle and Theorem Theorem have: we Then Theorem iesosadabtaycaatrsi L,Tm .] ntecs o case the In 0.1]. Thm. [L2, becomes: theorem characteristic Langer’s arbitrary and dimensions X o l ufcs hntesraei h rjciepae egive we plane, Fey projective of the method is the surface crit follow the effective We When an surfaces. surfaces. give all K3 to for on Feyzb [Br2] bundles example, [Br1] of For Bridgeland restrictions surfaces. by on developed bundles conditions of restrictions of study eutapisto applies result nescin.I h aeo ufc,Fenrstermbecom rem theorem Flenner’s to surface, bundles a of of restrictions sub case this for the in In criteria results give intersections. main The who curve. [MR] a Ramanathan to restricted when stable remain agrrcnl aeavr togrsrcintermwihhold which theorem restriction strong very a gave recently Langer oercn eeomnsi tblt odtosfrdrvdcatego derived for conditions stability in developments recent More oooo aeamr rcs etito hoe o ufcs[ surfaces for theorem restriction precise more a gave Bogomolov 2010 nti ae egv ucetciei o tbebnl nasmooth a on bundle stable a for criteria sufficient give we paper this In e od n phrases. and words Key uigtepeaaino hsatceteato a partia was author the article this of preparation the During and C TBLT ODTOSFRRSRCIN FVCO UDE ON BUNDLES VECTOR OF RESTRICTIONS FOR CONDITIONS STABILITY ⊂ C ahmtc ujc Classification. Subject Mathematics h ooooyo uhbnlsfrcre htlei h plan the in lie o that bundle re curves vector for when general bundles stable a such is of remain bundle cohomology to the vector surface the when projective criterion complex smooth a Abstract. X ⊂ (Flenner) (Bogomolov) (Langer) sasot uv fclass of curve smooth a is X r sagnrlcreo class of curve general a is ≥ any sn rdeadsaiiycniin,w iesffiin cr sufficient give we conditions, stability Bridgeland Using 2 . . and mohcremvn na ml ls.Fravco bundle vector a For class. ample an in moving curve smooth Let Let . ouisae fsevs tbevco ude,restricti bundles, vector stable sheaves, of spaces Moduli ( X C Let d ( ,H X, 1)( + ⊂ easot ufc and surface smooth a be ( X ,H X, 2 ) d eea uv fclass of curve general a easot oaie ufc.If surface. polarized smooth a be 2) + ) 2 easot oaie ufc.If surface. polarized smooth a be > d RJCIESURFACES PROJECTIVE dH rmr:1J0 46.Scnay 45,14D20. 14H50, Secondary: 14H60. 14J60, Primary: ∆( d > 1 + 2 E dH then , = ) max 1. r r 2 H > then , ONKOPPER JOHN Introduction 1 2 E · 2 | r c C max E 2 r 1 2 r ( H 4 r − 1 ssal if stable is | − l upre yNFRGgatDMS-1246844. grant RTG NSF by supported lly E C 2 − 1 ) 2 eyapedvsron divisor ample very a ssmsal if semistable is 2 1 , dH r 1 − 2 r 4 − then , ∆( ch ro izbuhsurfaces. Hirzebruch on or e h ln.A napiain ecompute we application, an As plane. the n H 1 2 2 E titdt uv.W ieastronger a give We curve. a to stricted es: r , ( ks F]ue h ahnr fstability of machinery the used [F1] akhsh 1 + 1 E 1 + ) o seas H].Ntby Bogomolov’s Notably, [HL]). also (see Bo] uhsrne tblt criterion. stability stronger much a E ) E . . trafrasal etrbnl on bundle vector stable a for iteria sa is | etaedet lne F]adMehta- and [Fl] Flenner to due are ject bks ogv etito theorem restriction a give to zbakhsh C E , r i tbeo iiosadcomplete and divisors on stable ain 2 o eyapedvsr nhigher in divisors ample very for s ishv e onwrslsi the in results new to led have ries n fvco bundles. vector of ons ssmsal if semistable is ∆( is ro ogaatetesaiiyof stability the guarantee to erion µ ufc ncaatrsi zero, characteristic in surface a f H µ E H smsal udeo rank of bundle -semistable 1 + ) ope rjciesraeto surface projective complex sal udeo rank of bundle -stable X . If . E let , E sa is µ H -semistable r on r on X 2 J. KOPPER
A notable difference between our first result and Langer’s theorem is that we are able to replace the requirement that C be ample with a weaker condition on the positivity of C. An application of this improve- ment is suggested by Proposition 3.10 in which we construct a (rational) map between moduli spaces: given a smooth polarized surface (X,H) and curve C X, under the conditions of Theorem 3.3, the restriction ⊂ map defines a morphism from the space of µH -stable sheaves on X to the space of stable sheaves on C. The precise statements of the main theorems are as follows:
Theorem (3.3). Suppose (X,H) is a smooth polarized surface, C an integral curve on X, and E a µH -stable vector bundle of rank r 2 on X. Then E C is stable if ≥ | C2 1 > r(r 1)∆(E)+ . 2H C − 2r(r 1)H2 · − Remark 1.1. During the preparation of this paper, S. Feyzbaksh pointed out that the proof of Theorem 3.3 relies only on the notion of tilt stability and a direct generalization therefore holds when restricting stable bundles to divisors on higher dimensional varieties. A proof of this fact in the case the divisor moves in the same ample class as the polarization can be found in Feyzbakhsh’s thesis [F2]. The author wants to thank S. Feyzbakhsh for her thoughtful conversations. In the special case that X is the plane and E is general in moduli, we obtain a much stronger bound for the degree of C that does not grow with the rank of E. Theorem (3.9). Suppose C P2 is a degree d integral curve and E is a general slope stable vector bundle 2 ⊂ on P with Chern character v such that ∆(v) 4. Then E C is stable if ≥ | d> 5 + 8∆(E)+5. Following Feyzbakhsh [F1], we prove both theoremsp using stability conditions for surfaces as constructed by Bridgeland [Br2], Arcara-Bertram [AB], and Toda [T]. If E is a slope stable sheaf on a surface X and i : C ֒ X a curve, then E and i E C fit into a distinguished triangle → ∗ | E i E C E( C)[1] E[1] → ∗ | → − → in the derived category b(X), where E( C)[1] denotes the complex associated to E( C) with the grading D − − shifted by one. If E and E( C)[1] are stable of the same phase, then i E C is semistable. We will show − ∗ | that in these circumstances E C is a slope semistable sheaf on C (Lemma 2.6). By slightly perturbing the | stability condition, we will show that E C is in fact stable (Lemma 2.7). The main content of the argument is producing| sufficient criteria to ensure that the conditions of Lemmas 2.6 and 2.7 hold. Specifically, we need E and E( C)[1] to be stable of the same phase. For a fixed Chern character v, there is a wall-and-chamber decomposition− of the stability manifold parametrizing stability conditions such that stable objects with Chern character v can only be destabilized when a wall is crossed [Br1]. To prove Theorem 3.3, we estimate the Gieseker wall, a wall in the stability manifold bounding the chamber consisting of stability conditions σ for which the only σ-semistable objects of Chern character v are the Gieseker semistable sheaves. Similarly, to prove Theorem 3.9 we estimate the effective wall for sheaves on P2. The effective wall bounds the chamber in which the general Gieseker semistable sheaf E is σ-semistable. The Giesker wall was computed explicitly in [CH1] and the effective wall for sheaves on P2 was computed in [CHW]. We expect the techniques used in proving Theorem 3.9 to generalize to any surface for which the effective wall is known. Structure of this paper. In Section 2 we recall necessary facts about stable sheaves and stability conditions for surfaces. In Section 3 we prove our restriction theorems. We conclude in Section 4 by giving some applications of restriction theorems to Brill-Noether problems for curves in the plane and on Hirzebruch surfaces. Acknowledgments. The author is grateful to Izzet Coskun, Takumi Murayama, Tim Ryan, and Matthew Woolf for their helpful feedback and input. The author is also grateful to the anonymous referees for their thoughtful responses.
2. Preliminaries In this section we will review facts about stable sheaves and stability conditions that will be used through- out the paper. All schemes are defined over the field of complex numbers. STABILITY CONDITIONS FOR RESTRICTIONS OF VECTOR BUNDLES ON PROJECTIVE SURFACES 3
2.1. Moduli of stable sheaves. For more details on moduli spaces of sheaves, we refer the reader to [HL]. Let (X,H) be a polarized variety and E a coherent sheaf on X of pure dimension d = dim(X). Then we may write d mi P (E,m)= χ(E X (m)) = αi(E) ⊗ O i! i=0 X for unique integers αi(E). The degree deg(E) is defined as the number αd−1(E) r αd−1( X ). If X is d−1 − · O smooth or an integral curve, then deg(E)= c1(E) H by Hirzebruch-Riemann-Roch [H, Ex. IV.1.9]. The reduced Hilbert polynomial of E is defined as· p(E,m) p(E,m)= . αd(E) We say that E is Gieseker semistable if for all proper subsheaves F E we have p(F,m) p(E,m) for m 0. We say E is Gieseker stable if strict inequality holds. ⊂ ≤ ≫ If E has rank r> 0, then we define the µH -slope of E to be the number deg(E) µ (E)= . H rHd We say that a torsion-free sheaf E is µH -semistable (or slope semistable) if µH (F ) µH (E) for all proper ≤ subsheaves F of rank strictly less than r. The sheaf E is said to be µH -stable (or slope stable) if strict inequality holds. Note that the notions of slope stability and Gieseker stability both depend on the choice of ample divisor H. d 1 Write c(E)=1+ c1(E)+ c2(E)+ cd(E)= i=1(1 + γi(E)), with γi(E) A (X) for all i. Then the Chern character of E is the polynomial··· ∈ Q d 1 2 ch(E)= exp(γi) = rk(E)+ c (E)+ (c (E) 2c (E)) + . 1 2 1 − 2 ··· i=1 X The degree n term of ch(E) is denoted chn(E). We extend the definition of the Chern character (and consequently that of the slope) to the derived category b(X) via D ∞ ch( E−1 E0 E1 )= ( 1)i ch Ei. ···→ → → → · · · − i=−∞ X Note that there are finitely many terms in this sum because we are working in the bounded derived category. Let E be an object in the derived category of X, not necessarily a sheaf. Write ch(E) = (ch0, ch1, ch2,..., chd). We will refer to ch0 as the rank of E though this number may be negative. If X is a surface, then define the discriminant of E to be the number 2 1 ch1 ch2 ∆(E)= 2 . 2 · ch0 − ch0 When working with Bridgeland stability conditions it is convenient to make a slight change of coordinates. For any Q-divisor D, define the twisted Chern character chD = exp( D) ch. Explicitly, − D2 chD = ch , chD = ch D ch , chD = ch D ch + ch . 0 0 1 1 − 0 2 2 − 1 2 0 We then define the twisted slope and twisted discriminant, respectively: D H ch1 D D · D ch0 (E) > 0 1 2 ch2 µH,D(E)= 2 , ∆H,D(E)= µH,D(E) . H ch0 2 D D 2 − H ch0 ch0 (E)=0 ∞ Note that µH,D(E) differs from µH (E)= µH,0(E) by a constant, and so E is a µH -stable (resp. semistable) sheaf if and only if µH,D(F ) < µH,D(E) (resp. µH,D(F ) µH,D(E)) for all D and all proper subsheaves ≤ F E with rank less than ch0(E). ⊂Finally, define the reduced twisted Hilbert polynomial of a positive rank sheaf E by
E χ(E X (mH D)) pH,D(m)= ⊗ O D − , ch0 (E) 4 J. KOPPER where the Euler characteristic is defined formally by Riemann-Roch. We say E is (H,D)-twisted Gieseker semistable if for every nonzero proper subsheaf F E, pF (m) pE (m) for m 0. We say E is ⊂ H,D ≤ H,D ≫ (H,D)-twisted Gieseker stable if strict inequality holds. 2 Lemma 2.1. Let E be a vector bundle. Then H ∆H,D(E) ∆(E), and there exists a Q-divisor D such 2 ≥ that H ∆H,D(E) = ∆(E). Proof. Write ch (E)= eH + ε and D = dH + δ, where H ε = H δ = 0. Then direct computation shows: 1 · · 2 1 2 H ∆H,D(E) ∆(E)= − 2 (ch0(E)δ ε) . − 2 ch0(E) −
This quantity is nonnegative by the Hodge index theorem and equals zero precisely when δ = ε/ ch0(E). 2 Note that we can take d = 0 in Lemma 2.1 to obtain a divisor D such that H ∆H,D(E) = ∆(E) and H D = 0. ·A famous theorem of Bogomolov says that if X is a smooth projective surface and E is slope semistable, then ∆(E) 0 [Bo]. By the above lemma, we also have ∆H,D(E) 0. For a given≥ Chern character v, Matsuki-Wentworth [MW] showed≥ that there are projective moduli spaces parametrizing S-equivalence classes of (H,D)-twisted semistable torsion-free sheaves with Chern character 2 v. We will denote this space by MX,(H,D)(v). When X = P we will always choose H to be the class of a line. In fact, if the Picard rank of X is one, then MX,(H,D)(v) does not depend on (H,D). We will supress the subscript X, (H,D) whenever it is clear from context. In order to compare sheaves on the surface X to their restrictions to a curve C X, we will need to know the Chern character of pushforwards of sheaves on C. The following standard lemma⊂ can be immediately verified in the case that C is smooth by using the Grothendieck-Riemann-Roch formula. Lemma 2.2. Let C be an integral curve on a smooth surface X, and let F be a sheaf of rank r on C. Then rC2 ch(i F )= 0, rC, deg(F ) . ∗ − 2 Proof. First we assume F = C . We then show the claim when F is any rank one sheaf, then we induct on O the rank r of F . If F = C, then we have the exact sequence O 0 X ( C) X i C 0, → O − → O → ∗ O → 2 .where i :֒ X is the inclusion. By additivity, ch(i∗ C ) = ch( X ) ch( X ( C)) = (0,C, C /2) as desired Suppose→ now the claim holds for (D) and p C Ois a smoothO point.− WeO have− the exact sequence− O ∈ 0 C (D p) C (D) Cp 0. → O − → O → → After pushing forward, we have 2 ch(i C (D p)) = ch(i C (D)) ch(i Cp)=(0,C, deg(D p) C /2). ∗ O − ∗ O − ∗ − − A similar argument shows that the claim holds for C (D + p). Since any line bundle can be obtained from O C by adding and subtracting smooth points, the claim must hold for any line bundle. O Now let F be any rank one sheaf. If L is an ample line bundle, then F L⊗m is globally generated for ⊗m ⊗ ⊗m m 0. In particular, F L has a global section s. The map C F L defined by multiplication by≫s is injective. Untwisting,⊗ we have an injection (L∨)⊗m F , andO → therefore⊗ an exact sequence → 0 M F Q 0 → → → → where M is a line bundle and Q is supported in dimension zero. Pushing forward, ch(i∗F ) = ch(i∗M)+ ch(i∗Q), and ch(i∗Q)=(0, 0,χ(Q)). All line bundles M have been shown to satisfy the claim, thus we have 2 ch(i∗F )=(0,C, deg(M)+ χ(Q) C /2). We show that deg(F ) = deg(−M)+ χ(Q). The Hilbert polynomial P (F,m) of F is given by P (F,m) = α1(F )m+α0(F ) for some coefficients α1, α0. Similarly, P (M,m)= α1(M)m+α0(M), P (Q,m)= α0(Q). By the additivity of P , α (F )= α (M)+α (Q) and α (F )= α (M). By definition, deg(F )= α (F ) α ( C )= 0 0 0 1 1 0 − 0 O deg(M)+ α0(Q). Since Q is supported at a point, α0(Q) = χ(Q). Thus the claim holds for any rank one sheaf F . Finally, we induct on r. If F is any sheaf of rank r> 1, then F fits into an exact sequence 0 M F Q 0 → → → → STABILITY CONDITIONS FOR RESTRICTIONS OF VECTOR BUNDLES ON PROJECTIVE SURFACES 5 where M is a line bundle and Q has rank r 1. By the induction hypothesis and the additivity of ch, the claim holds for F . − 2.2. Stability conditions on surfaces. Definition 2.3. [Br1] Let b(X) denote the bounded derived category of coherent sheaves on a projective D b variety X. A stability condition σ on (X) is a pair σ = (Z, ), where Z : K0(X) C is a group homomorphism and is the heart of a boundedD t-structure on b(AX) satisfying three properties:→ A D (1) (Positivity) Z maps nonzero objects in to the extended upper half-plane H = Reiθ : θ (0, π],R> 0 . A { ∈ (2) (Harder-Narasimhan} filtrations) For an object E of , define the σ-slope of E as A (Z(E)) µσ(E)= ℜ . − (Z(E)) ℑ We call E σ-stable (resp. semistable) if for every proper subobject F of E we have µσ(F ) <µσ(E) (resp. ). The pair ( ,Z) must satisfy the Harder-Narasimhan property: for every object E of , there is≤ a finite filtrationA A 0= E E En = E, 0 ⊂ 1 ⊂···⊂ such that Ei/Ei−1 is µσ-semistable, and µσ(Ei/Ei−1) >µσ(Ei+1/Ei) for all i. (3) (Support property) Fix a norm on Knum(X) R. Then there must exist a constant C > 0 such that k·k ⊗ E C Z(E) k k≤ k k for all semistable objects E in . A In the case when X is a smooth surface, Bridgeland [Br2], Arcara-Bertram [AB], and Toda [T] explicitly constructed stability conditions. Let H be an ample divisor on X and D any Q-divisor. For s R, define the following subcategories of Coh(X): ∈ ′ ′ s = Q Coh(X): Q is torsion or µH,D(Q ) >s for all quotients Q of Q Q { ∈ } ′ ′ s = F Coh(X): F is torsion-free and µH,D(F ) s for all subsheaves F of F . F { ∈ ≤ } b We then define the full subcategory s of (X) as A D • b −1 • 0 • i • s = F (X): (F ) s, (F ) s, (F )=0 for i = 1, 0 . A { ∈ D H ∈F H ∈Q H 6 − } Next, for E an object in b(X), s,t R, define. D ∈ 2 2 D+sH t H D+sH D+sH Zs,t(E)= ch (E)+ ch (E)+ iH ch (E). − 2 2 0 · 1 b Then the pair (Zs,t, s) defines a stability condition on (X) when t> 0 [AB], and if ch (E) > 0, we have: A D 0 2 2 (µH,D(E) s) t 2∆H,D(E) µσ(E)= − − − . µH,D(E) s − If E is a µH,D-stable torsion-free sheaf on X and s<µH,D(E), then E is in the category s. Similarly, the object E( C)[ 1] is given by the complex 0 E( C) 0 , concentratedA in degree 1, − − ··· → → − → → · · · − so E( C)[1] is in s if and only if E( C) is in s. Equivalently, we must have s µH,D(E( C)) = − A2 − F ≥ − µH,D(E) C H/H . − · Notation 2.4. If (X,H) is a smooth polarized surface and σ = (Zs,t, s) is a stability condition, then we A write the σ-slope as µs,t = µσ. We identify the family of stability conditions of the form (Zs,t, s) with the half-plane (s,t): s,t R,t> 0 called the (H,D)-slice. A { ∈ } An important feature of stability conditions is that for a fixed Chern character there is a wall-and-chamber decomposition of the space of stability conditions. A virtual wall in the (H,D)-slice is a set of points of the form W (E, F )= (s,t): µs,t(E)= µs,t(F ) { } b ′ If E is an object of (X) such that E is stable for σ = (Zs,t, s) but not for σ = (Zs′,t′ , s′ ), then there is a wall W (E, F ) separatingD the points (s,t) and (s′,t′) such thatA E is stable for nearby pointsA on one side 6 J. KOPPER of the wall but not for points on the other side. We call such walls actual walls. The actual walls in the (H,D)-slice are nested semicircles with bounded centers and radii [ABCH] (see [Mac] for the general case). b If E and F are objects of (X) of nonzero rank, then the wall W (E, F ) has center s0 and radius ρ0 given by: D
1 ∆H,D(E) ∆H,D(F ) 2 2 s0 = (µH,D(E)+ µH,D(F )) − ρ0 = (µH,D(E) s) 2∆H,D(E). 2 − µH,D(E) µH,D(F ) − − − The next lemma is standard. Lemma 2.5. Let (X,H) be a smooth polarized surface and E a coherent sheaf.
(1) If A E is a destabilizing subobject of E in s, then A is a sheaf. → A ′ ′ (2) If E[1] Q is a destabilizing quotient object of E[1] in s, then Q = Q [1] for some sheaf Q . → A Proof. Both statements easily follow from taking the long exact sequence in cohomology.
The method of [F1] is to consider the exact sequence 0 E( C) E i∗E C 0 where i : C ֒ X is an integral curve. There is a corresponding distinguished triangle→ − in →the derived→ | category:→ →
E i E C E( C)[1] E[1]. → ∗ | → − → If E and E( C)[1] are σ-stable of the same phase, then i E C is σ-semistable. The next lemma says that − ∗ | this is sufficient for E C to be slope semistable. | Lemma 2.6. Let (X,H) be a smooth polarized surface, i : C ֒ X an integral curve, and let F be a → torsion-free sheaf on C. Fix a stability condition σ = (Zs,t, s). If i∗F is σ-(semi)stable, then F is slope (semi)stable. A Proof. Applying Lemma 2.2, we obtain: ch (i F ) (D + sH) ch (i F ) µ (i F )= 2 ∗ − · 1 ∗ . s,t ∗ H ch (i F ) · 1 ∗ Let r denote the rank of F and let F ′ F be a subsheaf of rank r′. By Lemma 2.2, we have: ⊂ ′ 2 2 deg(F ′) r C (D + sH) r′C deg(F ) rC (D + sH) rC − 2 − · − 2 − · , H r′C ≤ H rC · · if F s σ-semistable, and strict inequality if F is σ-stable. The result follows immediately because every torsion sheaf is an object in s. A The following important consequence appears in [F1]. -(Corollary 2.7. Let (X,H) be a smooth, polarized surface, i : C ֒ X an integral curve, and E an (H,D → twisted stable sheaf on X. Let σ = (Zs,t, s) W (E, E( C)[1]) be a stability condition such that E and A ∈ − E( C)[1] are both σ-stable. Then E C is slope stable. − | Proof. Since σ lies on the wall W (E, E( C)[1]), we have that µs,t(i E C )= µs,t(E)= µs,t(E( C)[1]). By − ∗ | − Lemma 2.6, we see immediately that E C is slope semistable. By [Mar, Prop. 16], the condition σ may be ′ | ′ perturbed to some σ so that E, E( C)[1], and i∗E C are all σ -stable. The result now follows from Lemma 2.6. − |
3. Stable restrictions on smooth surfaces
Let X be a smooth surface and H an ample divisor on X. Suppose E is a µH,D-stable sheaf on X and C X an integral curve. The goal of this section is to give sufficient criteria for the restriction E C to be ⊂ | stable on C. To prove Theorem 3.3, we would like to apply Lemma 2.7. To do so, note that E and E C | fit into an exact sequence, 0 E( C) E i∗E C 0, and there is correspondingly a distinguished triangle in the derived category→ b(−X): → → | → D E i E C E( C)[1] E[1]. → ∗ | → − → STABILITY CONDITIONS FOR RESTRICTIONS OF VECTOR BUNDLES ON PROJECTIVE SURFACES 7
If we can find a stability condition σ = (Zs,t, s) such that E and E( C)[1] are σ-stable of the same slope, A − it will follow that E C is slope stable. The set of s,t such that E and E( C)[1] have the same slope is given by a semicircular wall| in the (H,D)-slice with center − C chD(E) C2 s = · 1 . ch (E)H C − 2H C 0 · · For any stable Chern character v, there exists a wall Wv, called the Giesker wall bounding the chamber consisting of stability conditions σ for which every (H,D)-twisted Gieseker stable sheaf of Chern character v is σ-stable. The Gieseker wall was computed in [CH1] for ∆H,D(E) 0. The method of proof of Theorem 3.3 may be thought of as giving a rough approximation of the Gieseker≫ wall. We will show that E cannot be destabilized by a subobject A unless the radius of W (A, E) is smaller than an explicit bound. Recall that by Lemma 2.5, any such destabilizing object must be a sheaf. Case (1) of the next lemma first appears in [ABCH, Lemma 6.3] for X = P2. The proof in the general situation is essentially identical. Lemma 3.1. Let E be an (H,D)-twisted Gieseker semistable sheaf on X of positive rank, and suppose A is a coherent sheaf. (1) If A E is a map of coherent sheaves which is an inclusion of σ-semistable objects of the same → slope for some σ = (Zs,t, s) in the (H,D)-slice with (s,t) W (A, E), then A and E are in s′ for all (s′,t′) W (A, E). A ∈ Q (2) If E Q∈is a map of coherent sheaves such that E[1] Q[1] is a surjection of σ-semistable objects → → of the same slope for some σ = (Zs,t, s) in the (H,D)-slice with (s,t) W (E[1],Q[1]), then E and ′ ′ A ∈ Q are in s′ for all (s ,t ) W (E[1],Q[1]). F ∈ We record a final technical lemma before proving the theorem. Lemma 3.2. Let E be a vector bundle of rank r on X and C any curve. Then there exists a line bundle L such that µH,D(E L) > 0 and ⊗ r(r 1)∆H,D(E L) µH,D(E L) r(r 1)∆H,D(E( C) L) µH,D(E( C) L). − ⊗ − ⊗ ≥ − − ⊗ − − ⊗ Proof. By the Hodge Index Theorem, we may write c1(E) = eH + ε,[C] = cH + γ, D = dH + δ, and ⊥ 2 2 2 2 c1L = lH + λ, where ε,γ,δ,λ H and ε ,δ ,γ , λ 0. By taking l 0, the inequality µH,D(E L) > 0 is immediately satisfied. ∈ ≤ ≫ ⊗ If γ = 0, then C is a multiple of H and we may take λ = 0. Suppose γ = 0. Lemma 2.1 gives: 6 2 1 2 H ∆H,D(E L) = ∆(E) (rδ ε rλ) ⊗ − 2r2 − − 2 1 2 H ∆H,D(E L ( C)) = ∆(E) (rδ ε rλ + rγ) ⊗ ⊗ O − − 2r2 − − Thus, 1 2 2γ ε ∆H,D(E L) ∆H,D(E L ( C)) = γ +2γ δ · 2γ λ . ⊗ − ⊗ ⊗ O − 2 · − r − · Since γ2 < 0, we may take L = lH + mγ for m 0 and l fixed. ≫ Theorem 3.3. Suppose (X,H) is a smooth polarized surface, C an integral curve on X, and E a µH,D-stable sheaf of rank r 2 on X. Then E C is stable if ≥ | C2 1 > r(r 1)∆(E)+ . 2H C − 2r(r 1)H2 · − Proof. Since twisting E by a line bundle does not affect its slope stability or the that of E C , by Lemma 3.2 | we may assume without loss of generality that µH,D(E) > 0 and
r(r 1)∆H,D(E) µH,D(E) r(r 1)∆H,D(E( C)[1]) µH,D(E( C)[1]). − − ≥ − − − − Take D as in Lemma 2.1 with H D = 0 and observe that the inequality in the statement of the theorem is then equivalent to · 2 D C ch1 (E) C 2 1 (3.3.1) · + µH,D(E) > r(r 1)H ∆H,D(E)+ . 2H C − rH C − 2r(r 1)H2 · · − 8 J. KOPPER
We show that if (3.3.1) holds, then E and E( C)[1] are σ-stable for stability conditions σ lying on W (E, E( C)[1]). Assume for a contradiction that− there is a torsion-free sheaf A and a map A E − → destabilizing E in s. There are two possibilities: ch (A) < r and ch (A) r. A 0 0 ≥ Assume first that ch0(A) < r. Let (s1, 0) and (s2, 0) be the endpoints of W (A, E) with s1 ′ 1 2 (3.3.2) s = µH,D(E) r(r 1)H ∆H,D(E) 0 − 2r(r 1)H2 − − − Suppose now that ch0(A) r and that W (A, E) is the largest actual wall in the (H,D)-slice where E is destabilized. Let ≥ 0 A E Q 0 → → → → be the destabilizing sequence in s with s<µH,D(E). We have that A and Q are semistable and consequently A D+sH that (Zs,t(A)), (Zs,t(Q)) > 0. Since (Zs,t)= H ch , we conclude that ℑ ℑ ℑ · 1 (3.3.3) H chD(A) > sH2 and H chD(Q) > sH2. · 1 · 1 There are two possibilities to consider: either (0,t) lies outside all actual walls for v for all t> 0 or there is a point (0,t) on W (A, E). In the former case, the wall W (A, E) is then bounded by the wall with center ′′ ∆H,D(E) (3.3.4) s0 = µH,D(E) . − µH,D(E) On the other hand, if (0,t) W (A, E) for some t, then A and Q must destabilize E in 0. By Inequality ∈ D D A D (3.3.3), we conclude that H ch1 (A) > 0 and H ch1 (Q) > 0. Since we have assumed that H ch1 (E) > 0, D · D · · it follows that 0 ′′′ 1 2 (3.3.5) s = µH,D(E) rH ∆H,D(E) 0 − 2rH2 − Comparing the centers in Equations (3.3.2), (3.3.4), and (3.3.5), we see that the center in Equation (3.3.2) corresponds to the largest wall. On the other hand, the center s of W (E, E( C)[1]) is − D 2 C ch1 (E) C 2 1 ′ s = · <µH,D(E) r(r 1)H ∆H,D(E) = s rH C − 2H C − − − 2r(r 1)H2 0 · · − by our assumption. It follows that if Inequality (3.3.1) holds, then E must be σ-stable for any stability condition σ = (Zs,t, s) with (s,t) W (E, E( C)[1]). We next show thatA E( C)[1] is∈σ-stable for− conditions σ on W (E, E( C)[1]). We repeat the above argument for E( C)[1] using− part (2) of Lemma 3.1 to show that if − − 2 D C ch1 (E) C 2 1 · + µH,D(E( C)) > r(r 1)H ∆H,D(E( C)) + , 2H C − rH C − − − 2r(r 1)H2 · · − then E( C)[1] is σ-semistable along W (E, E( C)[1]). This is automatically satisfied by our assumption. − − Thus E C is slope stable by Lemmas 2.6 and 2.7 if Inequality (3.3.1) is satisfied, which is equivalent to the inequality| in the theorem by our choice of D. STABILITY CONDITIONS FOR RESTRICTIONS OF VECTOR BUNDLES ON PROJECTIVE SURFACES 9 3.1. General sheaves on P2. Let v be a stable Chern character on P2 and C P2 any integral curve. ⊂ In this section we give sufficient criteria for the restriction E C of a general element of M(v) to be slope stable on C. The method of proof is similar to the above in| that we exploit the distinguished triangle E E C E( C)[1] E[1]. The main difference is that we require the wall W (E, E( C)[1]) to be outside→ the| → so-called− effective→ wall, beyond which the general Gieseker stable vector bundle is−σ-stable. The effective wall for MP2 (v) was computed in [CHW]. To bound the effective wall, we need a few definitions. For details on stable sheaves on P2 we refer to [LP] and [CHW]. An exceptional bundle E on P2 is a (Gieseker) stable vector bundle such that Ext1(E, E) = 0. A rational number ν is called an exceptional slope if it is the slope of an exceptional bundle. If ν is an exceptional slope, then there is a unique exceptional bundle of slope ν. Exceptional bundles are precisely the stable bundles E 1 with ∆(E) < 2 . Exceptional bundles are important for the classification of stable bundles on P2. If ν is an exceptional slope, let ∆ν denote the discriminant of the unique exceptional bundle of slope ν and put 3 √5+8∆ν x (ν)= − . 0 2 1 2 Let Iν be the open interval Iν = (ν x (ν),ν +x (ν)) and let P (x)= (x +3x+2) be the Hilbert polynomial − 0 0 2 of P2 . We define the function δ(µ) as O δ(µ)= P ( µ ν ) ∆ν , if µ Iν . −| − | − ∈ By [D], every slope µ of a sheaf lies in some interval Iν . A theorem of Dr´ezet and Le Potier [DLP] says that for a stable Chern character v with ∆(v) δ(µ(v)), the moduli space M(v) is a normal, irreducible, factorial projective variety of the expected dimension≥ r2(2∆(v) 1) + 1. Furthermore, if ∆(v) > δ(µ(v)), then Pic(M(v)) is a free abelian group of rank 2 [D]. We will assume− for the rest of this section that v is such that M(v) has Picard rank 2. Definition 3.4 ([CHW]). If v is a Chern character, we define the associated parabola Qv in the (µ, ∆)-plane as the equation P (µ + µ(v)) ∆(v) = ∆. − 1 v Theorem 3.5 ([CHW, Thm. 3.1]). The parabola Q intersects the line ∆= 2 at two points. If µ0(v) R 1 ∈ v is the larger of the two slopes such that (µ0(v), 2 ) Q , then there is a unique exceptional slope ν = ν(v) such that µ (v) (ν x (ν),ν + x (ν). ∈ 0 ∈ − 0 0 The slope µ0(v) is computed explicitly in [CHW] and is given by the formula 3 2µ(v)+ 5 + 8∆(v) µ0(v)= − − . 2 p The unique exceptional bundle from the previous theorem is called the corresponding exceptional bundle, and its Chern character is called the corresponding exceptional character. Definition 3.6. Given two coherent sheaves E and F on any smooth surface X, we define the pairing 2 χ(E, F )= ( 1)i exti(E, F ). − i=1 X By the Riemann-Roch formula, χ(E, F ) depends only the Chern characters of E and F and therefore also ∗ defines a pairing on Knum(X) R. If v is a Chern character, then v will denote the dual Chern character: if v = (r, c , ch ), then v∗ = (r,⊗ c , ch ). 1 2 − 1 2 Definition 3.7. Let F be a stable sheaf with Chern character v, and let w be the corresponding exceptional character. Define the corresponding orthogonal invariants µ+(v) and ∆+(v) as follows. (1) (µ+(v), ∆+(v)) = Qv Qw∗ if χ(v∗, w) > 0. (2) (µ+(v), ∆+(v)) = (µ(w∩), ∆(w)) if χ(v∗, w) = 0. (3) (µ+(v), ∆+(v)) = Qv Qw∗ if χ(v∗, w) < 0, where w∗ 3 = (rk(w), c (w) 3, ch (w)). ∩ −3 − − 1 − 2 10 J. KOPPER Let E M(v) be general, and suppose E+ is a sheaf with slope µ+(v), discriminant ∆+(v), and rank r+, where r+∈is sufficiently large and divisible. Then for stability conditions σ W (E, E+), there is an exact sequence ∈ 0 E+ Hom(E+, E) E W 0 → ⊗ → → → + + in the category s, where W is the mapping cone of the evaluation map E Hom(E , E) E. By computing the GiesekerA walls for E+ and W , we can show that E is σ-stable for σ⊗ W (E, E+), and→ thereby compute the effective wall for E. This is carried out in [CHW]. ∈ + By [CHW, Thm. 5.7], the center of the effective wall is s0 = µ (E) 3/2. To prove Theorem 3.9, it therefore suffices to show that the center of W (E, E( C)[1]) is at− most s −. The center of W (E, E( C)[1]) − 0 − is µ(E) d/2, where d is the degree of the curve C. To produce sufficient criteria to ensure µ(E) d/2 ∗ + Suppose now that χ(v , w) 0. Then µ lies in the interval [µ0 x0,µ0 + x0] (see [CHW]). Thus + ≤ − µ µ0 + x0, which immediately gives (2). Direct computation now shows that for ∆ 0, the right hand sides≤ of Inequalities (1) and (2) are dominated by ≫ 5 + 8∆(v) µ+(v) +1 µ(v). ≤ 2 − p Note that the assumption “∆(v) 0” in the above lemma is fairly mild. In fact, it suffices to have ∆(v) 4. To see this, observe that the≫ inequality ≥ 2∆ 3 √5+8∆ √5 √5+8∆ + − < +1 √1+8∆ 3 − 2 4 2 − holds as soon as 16∆6 32∆5 44∆4 56∆3 59∆2 20∆+4 > 0. The largest root of this polynomial is ∆ 3.3. Moreover,− the only− issue that− occurs− is that− the sum µ(v)+ µ(w) can tend to zero in the case χ(v∗,≈w) > 0, which in turn causes µ+(v) to grow very large. By [CHW, Lemma 3.6], µ(v)+ µ(w) is strictly positive when χ(v∗, w) > 0, and so in practice this computation can be made sharp. Theorem 3.9. Suppose C P2 is a degree d integral curve and E is a general slope stable vector bundle on 2 ⊂ P such that the moduli space MP2 (E) has Picard rank 2. Then E C is stable if d > 5 + 8∆(E)+5 and ∆ 4. | ≥ p Proof. Let v denote the Chern character of E. If d > 5 + 8∆(E)+5 and ∆ 4, then by the above lemma, µ(v) d/2 < µ+(v) 3/2. This shows that for the general E M(v), the≥ wall W (E, E( C)[1]) p lies outside the− effective− wall for−E. Arguing as in Theorem 3.3, E and E(∈ C)[1] are σ-stable for conditions− − lying on W (E, E( C)[1]). By Lemma 2.7, it follows that E C is stable. − | 3.2. Extending stable vector bundles. Rather than restricting from surfaces to curves, we can ask the opposite question: when does a stable vector bundle on C extend to a stable vector bundle on X? We will assume X is a smooth surface and v is a stable Chern character on X such that ∆(v) is large enough that MX,(H,D)(v) is well-behaved: Theorem (O’Grady [OG]). Let (X,H) be a smooth polarized surface, and v a Chern character with r(v) > 0. If ∆(v) 0, then MX,(H,D)(v) is normal, generically smooth, irreducible, and nonempty of the expected ≫ 2 2 dimension 2r ∆(v) (r 1)χ( X ). Furthermore, slope stable sheaves are dense in M (v). − − O X,(H,D) Let C X be a smooth curve in an ample class dA (with A not necessarily a multiple of the polarization ⊂ H) such that the restriction of a general element of MX,H (v) to C is stable. We have a restriction map MX,H (v) 99K UC(r, c1 C), where UC(r, e) denotes the moduli space parameterizing semistable rank r, degree e torsion-free sheaves on· C. The next proposition will show that for d 0, the restriction map is generically finite. Note that 2 ≫ 2 dim UC (r, e)= r (g 1)+1, where g is the genus of C, and dim M(v)= r (2∆(v) 1) + 1. By adjunction, 2 − − g grows with d , thus the image of MX,H (v) in UC(r, c C) is typically of large codimension. 1 · Proposition 3.10. With the above notation, if d 0, then the dimension of the image of the restriction ≫ map equals the dimension of MX,H (v). 2 Proof. Let M0 MX,H (v) denote the open, smooth subset of vector bundles E with Ext (E, E) = 0. Let ⊂ 1 E M0 be general and consider the differential TEM0 TE|C UC(r, c1 C). Since TEM0 = Ext (E, E) and ∈ 1 → 1 · 1 T UC(r, c C) = Ext (E C , E C), we have a map δ : Ext (E, E) Ext (E C , E C ). E|C 1 · C | | → C | | Next we apply the functor Hom(E, E ) to the short exact sequence 0 ( C) C 0. In ⊗− 1 → O − 1 →O→O → the associated long exact sequence in cohomology, there is a map Ext (E, E) Ext (E, E C ). Now, → | 1 1 ∨ 1 ∨ 1 Ext (E, E C )= H (X, E E C )= H (C, E C E C ) = Ext (E C , E C ). | ⊗ | | ⊗ | C | | 1 1 Thus the map Ext (E, E) Ext (E, E C ) in the long exact sequence is precisely the map δ (see, e.g., [CH2, Prop. 2.6]). Because E is stable,→ hom(E,| E( C)) = 0. Moreover, − 1 1 ∨ 1 ∨ ext (E, E( C)) = h (E E( C)) = h (E E ωX (dA)), − ⊗ − ⊗ ⊗ ⊗ O by Serre duality. Thus for d 0, this group vanishes and the map δ is injective. ≫ 12 J. KOPPER 4. Cohomology If E is a stable vector bundle on the smooth surface X and C X a curve, then we can use the standard exact sequence ⊂ 0 E( C) E i E C 0 → − → → ∗ | → to compute the cohomology of E C , at least when the cohomology of E is well-understood. The work of G¨ottsche-Hirschowitz on P2 [GH] and| Coskun-Huizenga on Hirzebruch surfaces [CH2] allows us to understand the cohomology of general elements of MX,(H,D)(v). After restricting to the curve C, we obtain some results that are interesting in the context of higher-rank Brill-Noether theory. In particular, we show that the number 0 h (C, E C ) can violate the expected dimension count of the Brill-Noether number. In light of Proposition | 3.10, we conclude that there is a fairly large-codimensional subspace of UC (r, e) consisting of vector bundles with unexpectedly many global sections. k k The Brill-Noether number, denoted ρr,e is the expected dimension of the subvariety Br,e of UC(r, e) consisting of rank r, degree e stable sheaves with at least k global sections. It is given by the following formula: k 2 ρ = r (g 1)+1 k(k e + r(g 1)) = dim UC(r, e) k(k χ(E)), r,e − − − − − − where g is the genus of C. 4.1. Plane curves. We will exploit the following result of G¨ottsche-Hirschowitz that describes the cohomol- ogy for a general stable sheaf. For this subsection, X = P2. Theorem (G¨ottsche-Hirschowitz [GH]). The general sheaf E M(v) of rank r 2 has at most one nonzero cohomology group: ∈ ≥ (1) If χ(E) 0 and µ(E) > 3, then H1(E)= H2(E)=0. (2) If χ(E) ≥ 0 and µ(E) −3, then H0(E)= H1(E)=0. (3) If χ(E) <≥ 0, then H0(E≤−)= H2(E)=0. We will make frequent use of the following Riemann-Roch calculation: Lemma 4.1. Suppose E is a rank r, degree e stable bundle on P2, and let C be a smooth curve of degree d. Then, (1) χ(E)= r +3e/2 + ch2(E). (2) χ(E( C)) = r + (3/2)(e dr) + ch (E) de + rd2/2. − − 2 − i Resolving E via the sequence 0 E( C) E i∗E C 0 we are led to consider nine cases: H (E) = 0, j → − → → | 0→ 1 6 H (E( C)) = 0 for i, j 0, 1, 2 . In all but two cases, H (E C )=0or H (E C ) = 0, and so there are no unexpected− global6 sections.∈{ The two} interesting cases are when |H0(E) = 0, H2(E| ( C)) = 0 and H1(E) = 0, H1(E( C)) = 0. In the latter case, a direct computation shows that these6 bundles− are Brill-Noether6 general6 when deg(− C)6 is large: Proposition 4.2. Suppose E is a general element of M(v) such that H1(E) =0 and H1(E( C)) =0 for 0 k 6 − 6 a smooth degree d curve C. Let e = deg(E) and k = h (E C ). Then ρ 0 for d 0. | r,de ≥ ≫ 0 1 1 1 Proof. From the long exact sequence in cohomology we see that h (E C ) h (E( C)) and h (E C ) h (E). Thus, | ≤ − | ≤ 3e 3 rd2 ρk r2(g 1)+1+ r + + ch (E) r (e dr) ch (E)+ de r,de ≥ − 2 2 − − 2 − − 2 − 2 (d 1)(d 2) 3e 3 rd2 = r2 − − 1 +1+ r + + ch (E) r (e dr) ch (E)+ de 2 − 2 2 − − 2 − − 2 − 2 d2 er = − ch (E) + P (d), 2 2 − 2 where P (d) is a polynomial of degree one in d. Since er/2 ch2(E) = χ(E)+ r and r χ(E) > 0, we see that for d 0 we must have ρk 0. − − − − ≫ r,de ≥ For the other case, we have: STABILITY CONDITIONS FOR RESTRICTIONS OF VECTOR BUNDLES ON PROJECTIVE SURFACES 13 Proposition 4.3. Let C be a smooth plane curve of degree d, and suppose E is a general element of M(v) 0 2 0 such that H (E) =0 and H (E( C)) =0. Suppose further that χ(E) > r. Let e = deg(E) and k = h (E C ). Then ρk < 0 but6 Bk is nonempty− for6 d 0. | r,de r,de ≫ k Proof. By the long exact sequence in cohomology, it follows that ρr,de < 0 if 3e 3e r2(g 1)+1 < r + + ch (E) r + + ch (E) de + r(g 1) . − 2 2 2 2 − − Since g = (d 1)(d 2)/2, the left-hand side is dominated by the term d2r2/2 and the right-hand side is − − 2 2 2 dominated by the term d (r /2+3er/4+ r ch2(E)/2). Thus the inequality holds for large d if r /2+3er/4+ 2 r ch2(E)/2 > r /2. This is equivalent to the inequality χ(E) > r. Note that µ(E( C)) = µ(E) d, so the − − k hypotheses are preserved by taking d large. The restriction E C is semistable for d 0, thus the locus Br,de is nonempty. | ≫ 0 Note that if r and ch2(E) are fixed, then h (E) grows with e and thus the hypotheses of the above theorem hold for fixed ch (E) with d 0 and e 0. 2 ≫ ≫ 4.2. Curves on Hirzebruch surfaces. Coskun-Huizenga [CH2] computed the cohomology of a general stable sheaf on a Hirzebruch surface. Let X = Fm be the Hirzebruch surface X = P( P1 P1 (m)), m> 0. Let M denote the curve class of self-intersection m and F the class of a fiber. If v =O (r,⊕ c O, ch ) is a stable − 1 2 Chern character on X, then define v(v) = c1/r. The cohomology of the general element E MX,(H,D)(v) can be determined with the intersection numbers v(v) M and v(v) F in addition to the Euler∈ characteristic χ(v). · · Theorem 4.4 ([CH2, Theorem 3.1]). Let v be a stable Chern character on Fm with positive rank and E M (v) a general stable sheaf. Then ∈ X,(H,D) (1) If v(v) F 1, then h2(E)=0. (2) If v(v) · F ≥−1, then h0(E)=0. (3) If v(v) · F ≤−= 1, then h1(E)= χ(v) and all other cohomology vanishes. · − − Suppose now v(v) F > 1. Then either of the numbers h0(E) or h1(E) determines the Betti numbers of E: · − (4) If v(v) M 1, then E has at most one nonzero cohomology group. If χ(v) 0 then h0(E)= χ(v), and if χ· (v)≥−0, then h1(E)= χ(v). ≥ ≤ − (5) If v(v) M < 1, then H0(E) = H0(E( M)). · − ∼ − If v(v) F < 1 and r 2, then the cohomology of E can be determined by Serre duality. · − ≥ Lemma 4.5. Let X = Fm. Then: 1 (1) KX = 2M (e+2)F , and if C is a smooth curve of class aM +bF , then g(C)= 2 (1 a)(am 2b+2). (2) Let H =− aM−+ bF . Then H is ample if and only if a> 0 and b>am. If H is ample,− C a− curve of class dH, and E a µH -stable sheaf of rank r 2, then E C is semistable if ≥ | 1 d> 2r(r 1)∆(E)+ . − 2r(r 1)(ab a2m) − − Proof. Statement (1) is a direct adjunction computation, and statement (2) is Theorem 3.3. 2 0 In the same spirit as the P case, we are interested in understanding the asymptotic behavior of h (E C ) when a and when b . Note that we have the following: | → ∞ → ∞ v(E( C)) M v(E( C)) M − · → ∞ as a , − · → −∞ as b χ(E( C)) → ∞ χ(E( C)) → ∞ − → −∞) − → ∞ ) 2 0 1 As with P , the interesting cases occur when E C has both H and H . Because of the behavior of the Betti numbers of E( C) as a and b grow independently,| there are two possibilities: Either E has only H0 and E( C) has both−H1 and H2, or E and E( C) have only H1. The latter case can only occur when [C]= adM− + bdF with v(E( C)) M 1, χ(E−( C)) < 0, and a + b 0. As with the analogous case on P2, this does not produce any− stable· bundles≥− on C−with unexpectedly≫ many global sections. On the other hand, we have: 14 J. KOPPER Proposition 4.6. Let H = aM + bF be ample, and let E be a general element of MX,(H,D)(v) for some stable Chern character v such that v(v) F < 0 and E has only H0. Let C be a smooth curve of class dH, 0 · k k and put k = h (C, E C ), e = deg(E C ). If χ(E) > r and a 2, then ρr,e < 0, but Br,e is nonempty for b, d 0. | | ≥ ≫ Proof. Write c1(E) = xM + yF . Then x < 0 by assumption and v(E( C)) F = x/r ad < 1. Thus 0 2 0 0 − · − − h (E( C)) = 0. Since h (E C ) = 0, we have h (E C ) h (E)= χ(E). It therefore suffices to show − | | ≥ (4.6.1) r2 (g 1))+1 <χ(E) (χ(E) e + rg) . − − 1 Since g = 2 (1 ad)(adm 2db + 2), we see that g as b . Then e = d( amx + bx + ay) with x < 0. Since we− have also− assumed that χ(E) > r, the→ right ∞ side→ dominates ∞ the left− for b 0. Further, as b grows, v(E( C)) F is unchanged and h0(E( C)) = 0. We may therefore find b sufficiently≫ large so that − · − Inequality (4.6.1) holds for all d. Since E C is semistable for large d, the claim follows. | References [AB] D. Arcara and A. Bertram, Bridgeland-stable moduli spaces for K-trivial surfaces. J. Eur. Math. Soc. 15 no. 1 (2013), 1–38. [ABCH] D. Arcara, A. Bertram, I. Coskun, and J. Huizenga, The minimal model program for the Hilbert scheme of points on P2 and Bridgeland stability. Adv. Math. 235 (2013), 580–626. [Bo] F. A. Bogomolov, Stable vector bundles on projective surfaces. Sb. Math. 81 no. 2 (1995). [BHL+] B. Bolognese, J. Huizenga, Y. Lin, E. Riedl, B. Schmidt, M. Woolf, and X. Zhao, Nef cones of Hilbert schemes of points on surfaces. Algebra Number Theory 10 no. 4 (2016), 907–930. [Br1] T. Bridgeland, Stability conditions on triangulated categories. Ann. of Math. (2) 166 no. 2 (2007), 317–345. [Br2] T. Bridgeland, Stability conditions on K3 surfaces. Duke Math. J. 141 no. 2 (2008), 241–291. [CH1] I. Coskun and J. Huizenga, The nef cone of the moduli space of sheaves and strong Bogomolov inequalities. Israel J. Math., to appear. [CH2] I. Coskun and J. Huizenga, Brill-Noether theorems and globally generated vector bundles on Hirzebruch surfaces. arXiv:1705.08460. [CHW] I. Coskun, J. Huizenga, and M. Woolf, The effective cone of moduli spaces of sheaves on the plane. J. Eur. Math. Soc. 19 no. 5 (2017), 1421–1467. [D] J.-M. Dr´ezet, Fibr´es exceptionnels et vari´et´es de modules de faisceaux semi-stables sur P2(C). J. Reine Angew. Math. 380 (1987), 14–58. [DLP] J.-M. Dr´ezet and J. Le Potier, Fibr´es stables et fibr´es exceptionnels sur P2. Ann. Sci. Ec.´ Norm. Sup´er. (4) 18 no. 2 (1985), 193–243. [F1] S. Feyzbakhsh, Stability of restrictions of Lazarsfeld-Mukai bundles via wall-crossing and Mercat’s conjecture. arXiv:1608.07825. [F2] S. Feyzbakhsh, Bridgeland Stability conditions, stability of the restricted bundle, Brill-Noether theory, and Mukai’s program. Ph.D. thesis (2018), University of Edinburgh. [Fl] H. Flenner, Restrictions of semistable bundles on projective varieties. Comment. Math. Helv. 59 (1984), 635–650. [GH] L. G¨ottsche and A. Hirschowitz, Weak Brill-Noether for vector bundles on the projective plane, in Algebraic Geometry (Catania, 1993/Barcelona, 1994), 6374, Lecture Notes in Pure and Appl. Math., 200 Dekker, New York. [H] R. Hartshorne, Algebraic Geometry, Springer-Verlag, New York, 1977. Graduate Texts in Mathematics, no. 52. [HL] D. Huybrechts and M. Lehn, The geometry of moduli spaces of sheaves, second edition, Cambridge Mathematical Library, Cambridge Univ. Press, Cambridge, 2010. [L1] A. Langer, Semistable sheaves in positive characteristic. Ann. of Math. (2) 159 (2004), 251–276. [L2] A. Langer, A note on restriction theorems for semistable sheaves. Math. Res. Lett. 17 no. 5 (2010), 823–832. [LP] J. Le Potier, Lectures on Vector Bundles, translated by A. Maciocia, Cambridge Studies in Advanced Mathematics, 54, Cambridge Univ. Press, Cambridge, 1997. [Mac] A. Maciocia, Computing the walls associated to Bridgeland stability conditions on projective surfaces. Asian J. Math. 18 no. 2 (2014), 263–279. [Mar] C. Martinez, Duality, Bridgeland wall-crossing and flips of secant varieties. Internat. J. Math. 28 no. 2 (2017), 1750011, 40. [MR] V. Mehta and A. Ramanathan, Semistable sheaves on projective varieties and their restriction to curves. Math. Ann. 258 no. 3 (1982), 213–224. [MW] K. Matsuki and R. Wentworth, Mumford-Thaddeus principle on the moduli space of vector bundles on an algebraic surface, Internat. J. Math., 8 no. 1 (1997), 97–148. [OG] K. O’Grady, Moduli of vector bundles on projective surfaces: Some basic results. Invent. Math. 123 (1996), 141–207. [T] Y. Toda, Stability conditions and extremal contractions. Math. Ann. 357 no. 2 (2013), 631–685. Department of Mathematics, The Pennsylvania State University, University Park, PA, 16802 E-mail address: [email protected]